Barts CRL dataset

This dataset is part of a technical validation study for a deep learning model designed to detect, segment, and classify liver lesions in colorectal cancer (CRC) patients using staging CT scans. The dataset includes anonymized CT images from colorectal cancer staging, gathered retrospectively from three large hospitals in London: St Bartholomew’s Hospital, The Royal London Hospital, and Newham University Hospital, covering the period between January 2014 and December 2019.

Purpose: The goal of this dataset is to support the development and validation of deep learning algorithms for detecting liver metastases and distinguishing between benign and malignant liver lesions in CRC patients. The dataset includes detailed lesion-level annotations for validation and evaluation of lesion detection, segmentation, and classification algorithms.

Contents:

CT Scans (filename: images.zip): A total of 220 anonymized contrast-enhanced CT scans of CRC patients.

Ground Truth Annotations (filename: lms.zip): Liver lesions (both benign and malignant) are manually segmented and classified based on confirmatory imaging, including MRI or PET scans, or long-term follow-up. All annotations were done by a technologist and confirmed by a radiologist. Labels:Background : 0Benign lesions: 2Malignant lesions: 3

Lesion Classification: Lesions are categorized as benign or malignant based on secondary modality findings or growth characteristics on follow-up imaging.